Qué es el data poisoning, la técnica de los artistas para “envenenar” inteligencias artificiales generativas de imágenes

The central tension in generative artificial intelligence lies in the legal and ethical controversy surrounding the training data used by developers. Major tech firms rely on vast amounts of human-created content, including copyrighted works, to train their algorithms. This practice has triggered significant legal battles, with authors and agencies suing companies for alleged intellectual property violations, arguing that the unauthorized extraction of protected material constitutes a direct infringement of creators' rights. In response to these perceived injustices, a new defensive strategy has emerged involving "data poisoning" techniques like Nightshade. By making imperceptible alterations to artistic works, creators can manipulate how AI models interpret their content, causing the algorithms to learn incorrect associations. This approach aims to disrupt the accuracy and reliability of generative models, forcing a critical reevaluation of how these systems process and utilize digital assets without explicit permission. This development is highly relevant to open data because it challenges the assumption that publicly available information is free from ethical constraints. It highlights the urgent need for transparent data governance and clear licensing frameworks that distinguish between open access and restricted intellectual property. Ultimately, it underscores the importance of respecting creator rights within open ecosystems, pushing for a more equitable balance between technological innovation and individual ownership in the data-driven economy.

Source: tn.com.ar
Published on 2024-01-01